After major conferences from Salesforce and HubSpot this week, it’s clear that major companies are getting on board with AI agents. Join Mike and Paul as they chat about the ups and downs of these AI agents and what they mean for the future of work. And that’s not all—our hosts also talk about Runway’s latest partnership with Lionsgate, Sam Altman’s “Goal 3,” LinkedIn’s training on user data, and more.
Listen or watch below—and see below for show notes and the transcript.
Listen Now
Watch the Video
Timestamps
00:03:10 — Salesforce, HubSpot AI Agent Announcements & AI Agents in the Enterprise
- Salesforce
- HubSpot
- AI Agents in the Enterprise
00:29:28 — Runway + Lionsgate
00:34:36 — BlackRock AI Infrastructure
00:38:45 — Sam Altman: “Goal 3”
00:42:24 — YouTube + AI
00:46:31 — LinkedIn Training on User Data
00:49:43 — NotebookLM Update
00:52:10 — Plex(dot)it
00:53:56 — Introducing the ExecAI Newsletter
Summary
Salesforce and HubSpot
Both Salesforce and HubSpot had major conferences this past week where they dropped a ton of AI updates…
First, at this year’s Dreamforce, Salesforce continued its trend of going all-in on AI agents.
The biggest announcement was Salesforce’s Agentforce platform. The company bills Agentforce as a “customizable AI agent builder,” with an agent being defined by Salesforce as “proactive, autonomous application that provides specialized, always-on support to employees or customers.
Not to be outdone, HubSpot’s annual INBOUND conference also delivered the AI goods.
Far and away the biggest announcement was HubSpot Breeze, what HubSpot is calling its “AI engine,” or a series of copilots and agents across its platform to improve how you work with any HubSpot product.
HubSpot says Breeze has over 80 AI-powered features across every HubSpot product.
AI Agents in the Enterprise
AI agents are on the brink of shaking up the enterprise world. We are starting to see some exciting experimental demos that show off how these advanced systems can handle tasks with human-like accuracy.
Roetzer’s prediction on the real breakthrough of these agents is expected in 2025 when AI agents are set to reach new heights in reliability and autonomy.
As these powerful AI tools start integrating into our workplaces, the disruption to knowledge work will start to become more tangible and measurable.
Runway and Lionsgate
AI video generation company Runway has inked a groundbreaking deal with Lionsgate, the studio behind popular franchises like “John Wick” and “Twilight.”
This partnership marks the first public collaboration between a generative AI startup and a major Hollywood studio.
Under this agreement, Runway will train a custom video model using Lionsgate’s extensive movie catalog. The studio’s creative talent, including filmmakers and directors, will gain access to this AI model to enhance their work.
Lionsgate vice chair Michael Burns emphasized the potential for this technology to augment the creative process.
Runway is also exploring ways to license these models as templates, allowing individual creators to build and train their own custom models.
This week’s episode is brought to you by MAICON, our 5th annual Marketing AI Conference that took place last week, Sept. 10-12 in Cleveland, Sept. 10 – 12. On-demand recordings are now available for purchase.
Read the Transcription
Disclaimer: This transcription was written by AI, thanks to Descript, and has not been edited for content.
[00:00:00] Paul Roetzer: Hubspot, Salesforce, Google, Microsoft, none of them are going to talk about these agents being replacements to workers. They’re all going to talk about them being
[00:00:09] Paul Roetzer: unlocks and, you know, enhancements and augmentations. So let’s get one thing clear, like agents will replace people. This is regardless
[00:00:18] Paul Roetzer: of the marketing messaging that people use, they will, there will be wonderful things that come from these agents, but they’ll take jobs too.
[00:00:25] Paul Roetzer: Welcome to the Artificial Intelligence Show, the podcast that helps your business grow smarter by making AI approachable and actiohighly nable. My name is Paul Roetzer. I’m the founder and CEO of Marketing AI Institute, and I’m your host. Each week, I’m joined by my co host. and Marketing AI Institute Chief Content Officer, Mike Kaput, as we break down all the AI news that matters and give you insights and perspectives that you can use to advance your company and your career.
[00:00:55] Paul Roetzer: Join us as we accelerate AI literacy for all. [00:01:00] WElcome to episode 116 of the artificial Intelligence Show. I’m your host, paul Roetzer, along with my co host, Mike Kaput. We are, was MAICON just last week, Mike? Or is that two weeks
[00:01:14] Mike Kaput: It was two weeks ago, but honestly, I didn’t know the answer to that question.
[00:01:19] Paul Roetzer: I feel like my life has just like been in slow motion since then. Oh no, I was in Seattle
[00:01:24] Mike Kaput: yeah, because we were traveling last week, which felt like a continuation of MAICON.
[00:01:29] Paul Roetzer: I, You know, I had this grand plan of like taking a week off and just like decompressing and then, yeah, then I left for Seattle and you went to New York, right? Like, yes. Okay. Yeah. So, so here we go. We are, we are back in our home studios to record
[00:01:44] Paul Roetzer: episode 116. But this episode is brought to us by MAICON OnDemand. So if you missed MAICON, which was September 10th to 12th, I could have just looked at the CTA read here, which was two weeks ago. If you weren’t able to join us, you can actually rewatch [00:02:00] the sessions or watch the sessions for the first
[00:02:01] Paul Roetzer: time or rewatch your favorite sessions if you were there.
[00:02:04] Paul Roetzer: there’s over 25 sessions available on demand, including all
[00:02:08] Paul Roetzer: the main stage keynotes.
[00:02:09] Paul Roetzer: As well as the featured breakout sessions and some of the extremely popular, and highly rated panels. Some amazing stuff including like the ai, copyright intellectual property panel. Mike’s ai, 30 AI tools in 30 minutes talk.
[00:02:24] Paul Roetzer: My opening, wrote to Ai g to a GI keynote. Just a ton of stuff packed in. So you can get those now. They are up and live. You can go to macon.ai. That’s M-A-I-C-O n.ai. And click buy MAICON 2024 on demand. If you’re so inclined, you can also register for MAICON 2025. I am shocked by how many people have already
[00:02:46] Paul Roetzer: registered for next year’s event.
[00:02:48] Paul Roetzer: It’s every day people are registering,
[00:02:50] Paul Roetzer: which is pretty remarkable. So, again, MAICON. ai if you want to check that
[00:02:55] out. And,
[00:02:57] Paul Roetzer: Yeah, we are going to talk about AI [00:03:00] agents today, and we are not the only ones. It was the hot topic last week at
[00:03:04] Paul Roetzer: Dreamforce and InBound. So Mike, let’s kick this off with a little AI agent conversation.
[00:03:10] Salesforce, HubSpot AI Agent Announcements & AI Agents in the Enterprise
[00:03:10] Mike Kaput: Sounds good, Paul.
[00:03:11] Mike Kaput: So we are going to kind of tee this up a little differently. We’re first going to, as our first main topic, start off talking about some of the announcements that came out of Dreamforce and InBound. As you alluded to, they have a lot to do with agents. And
[00:03:25] Mike Kaput: then we’re going to seamlessly transition into Paul getting your kind of thoughts
[00:03:29] Mike Kaput: on, Overall, the trend of aI agents.
[00:03:32] Mike Kaput: Especially in the enterprise,
[00:03:33] Mike Kaput: like, where are we at? Where are we going? What do we need to know as we’re getting kind of bombarded
with all these new products and new hype around AI agents? So, both Salesforce and HubSpot had major conferences this past week at this year’s Dreamforce, which is Salesforce’s conference.
[00:03:52] Mike Kaput: They continued Their recent trend of going all in on AI agents.So the biggest announcement, which we’ve talked about [00:04:00] on previous episodes as it was being teased out before the event was its agent force platform. So Salesforce builds agent force as a quote, customizable AI agent builder, and they kind ofdefine an agent.as
[00:04:15] Mike Kaput: a quote proactive autonomous application that provides specialized always on support to employees or customers. They’re equipped with the
[00:04:22] Mike Kaput: necessary business knowledge to execute tasks according to their specific role. Now, AgentForce, in addition to letting you build your own
[00:04:31] Mike Kaput: agents on top of Salesforce products, also has some pre built agents to get you started. These include a service agent, an SDR
[00:04:40] Mike Kaput: agent, and even a sales coach agent. Now there were tons and tons of other AI and product focused announcements at, Dreamforce, but the company did highlight notably that Slack is basically turning into
[00:04:55] Mike Kaput: an AI agent hub. You’re going to be able to interact with and guide AI [00:05:00] agents to do work for you right from Slack.
[00:05:03] Mike Kaput: Salesforce is expanding its investment in AI through Salesforce Ventures up to
[00:05:08] Mike Kaput: a billion dollars. So they’re increasing that dramatically. And the company is also offering what they call free hands on AI courses and certifications through 2025.
[00:05:19] Mike Kaput: This is an investment they say has cost them more than 50 million dollars to train
[00:05:24] Mike Kaput: people for quote, the jobs of tomorrow. Now, not to be outdone HubSpot’s annual inbound conference also delivered the AI goods. The biggest announcement here was HubSpot Breeze.
[00:05:36] Mike Kaput: This is what HubSpot is calling its quote AI engine, and it’s basically a series of features, co pilots,
[00:05:43] Mike Kaput: and agents built into its different hubs to improve how you work with HubSpot products. HubSpot says Breeze has over 80 AI powered features across every platform. HubSpot product, and these include Breeze AI agents and co pilots that do
[00:05:58] Mike Kaput: things for you across social, [00:06:00] content, prospecting, customer service, and more. For instance, there’s a Breeze content agent that will automatically write case studies for you, based on
[00:06:08] Mike Kaput: transcripts and notes that you give it. There is something called Breeze Intelligence, which is an AI enhanced data set
[00:06:15] Mike Kaput: for enriching your CRM records. There are a bunch of AI features augmenting the marketing and content hubs, including, interestingly, a video remix feature, which repurposes existing videos into short clips, and actually uses a an integration with Haygen to generate new
[00:06:34] Mike Kaput: video using AI. And there’s the ability to use Breeze to quickly and easily understand
[00:06:40] Mike Kaput: everything about your data.
[00:06:41] Mike Kaput: It’ll tell you everything from prospect intent, to generate report summaries with AI, to tell you more about user behavior on the site.
[00:06:50] Mike Kaput: Now, last but certainly not least, CTO and co founder Dharmesh Shah, in his keynote
[00:06:57] Mike Kaput: at the event, talked all about something called AI. [00:07:00] Agent.AI
[00:07:01] Mike Kaput: ai. Now this is a new website that Dharmesh and team have launched and I’m just going to read straight from the website their description of kind of what this is. and what’s going
[00:07:11] Mike Kaput: on here.So they literally have a section on the site saying, what is this? And you click on it and it says you might be wondering what is this? Or maybe wondering why is this? Let’s start with what this is.
[00:07:22] Mike Kaput: Agent. Ai is a project that I, Dharmesh, co founder
[00:07:25] Mike Kaput: CTO at HubSpot, have been
[00:07:27] Mike Kaput: working on for a while. It’s a place where you can find and connect with AI agents. These agents can do all sorts
[00:07:33] Mike Kaput: of things. They can help you GSD, get stuff done.
[00:07:36] Mike Kaput: Right now, the agents are available, are internally developed, but the plan is to open this up to the world. Now on to the
[00:07:44] Mike Kaput: why. Why am I doing this? Because I’ve been writing software for 30 plus years and never have I
[00:07:49] Mike Kaput: been more excited about the potential for AI, particularly AI agents. In a year or two, I think it’ll be, as common to build AI agents to solve customer [00:08:00] problems as
[00:08:00] Mike Kaput: it is to build web apps today. And indeed, if you go to agent. ai right now, you can find all sorts of agents. The professional network for AI agents is what
[00:08:10] Mike Kaput: they build at the moment. And you basically can go hire an AI agent, which is using an AI agent to, for instance, generate images or memes, which
[00:08:20] Mike Kaput: are two of the available ones that you can hire.
[00:08:22] Mike Kaput: on the website. So, Paul, that’s kind of what’s going on at Salesforce and HubSpot, but I like
[00:08:29] Mike Kaput: Dharmesh’s final comment there as a transition into kind of the bigger second topic today, which is generally AI agents in the
[00:08:37] Mike Kaput: enterprise, like with all this buzz around literally Salesforce is all in on agents, It seems like HubSpot is heavily in on agents, where are we actually at today with AI agents?
[00:08:50] Mike Kaput: Like, are these all suddenly ready? to scale in the enterprise?
[00:08:53] Mike Kaput: Like, is there real value here yet?
[00:08:55] Paul Roetzer: Yeah, so I found myself asking these same questions last week. [00:09:00] So interestingly, I think on the podcast last week, we touched on this idea that Agents are sort of taking on different definitions that it’s not really clear what exactly everyone’s talking
[00:09:11] Paul Roetzer: about, because if we go back, like, to episode 87, the AI timeline, where I laid out that AI agents were going to take off, like, 2025 to 2027, There was a certain kind of agent we were thinking about, and we were talking about, and that
[00:09:26] Paul Roetzer: was kind of more the generally capable personal agent. So, just to recap, like, in that timeline, what I said was agents that can take action in 2025 are
[00:09:36] Paul Roetzer: happening, but they’re mostly experimentations and demonstrations. And yet we have. Salesforce and HubSpot touting
[00:09:43] Paul Roetzer: that these things are everywhere and they’re ready to go.
[00:09:46] Paul Roetzer: at the time, what I said was lots of manual work to get the agents to function reliably, plus lots of human oversight. So these
[00:09:53] Paul Roetzer: aren’t like autonomous agents, is what my belief was. And again, keep in mind, we recorded Episode 87 in March
[00:09:59] Paul Roetzer: of this year. [00:10:00] So this was me looking out ahead and saying, here’s what
[00:10:02] Paul Roetzer: I think is going to happen in 2024.
[00:10:04] Paul Roetzer: And then I basically said that companies generally won’t be willing to give up the data and privacy needed to get these benefits.
[00:10:10] Paul Roetzer: So while a lot of small, mid sized businesses may jump on the agent train, it’s, it’s going to be a lot of enterprises that are
[00:10:17] Paul Roetzer: going to say, no way, like it needs access to all this data that we’re not going to give it
[00:10:20] Paul Roetzer: access to.now that might not happen within like a Salesforce environment, if you already trust
[00:10:25] Paul Roetzer: Salesforce and it has access to the data for like a customer service use case or a sales use case.
[00:10:30] Paul Roetzer: then you may be willing to use those agents to do those kinds of things. And then I said, starting in 2025, AI can now take actions more reliably with, less human oversight, and
[00:10:41] Paul Roetzer: then it starts to take off in selected domains and verticals initially, and then more generally over time. Early instances of full autonomy start
[00:10:51] Paul Roetzer: to emerge in 2025, meaning you give it a goal and it goes and does things.
[00:10:55] Paul Roetzer: The thing for you with no human oversight and then disruption to knowledge [00:11:00] work starts to become more tangible and measurable. Now HubSpot,
[00:11:03] Paul Roetzer: Salesforce, Google, Microsoft, none of them are going to talk
[00:11:07] Paul Roetzer: about these agents being replacements to workers. They’re all going to talk about them being unlocks and, you know, enhancements and augmentations.
[00:11:15] Paul Roetzer: They all have to say that. So let’s get one thing. clear, like agents will replace people. This is regardless
[00:11:23] Paul Roetzer: of the marketing messaging that people use, um, It will augment them. It will
[00:11:28] Paul Roetzer: enhance people’s jobs. They will, there will be wonderful things that come from these agents, but they’ll take jobs too.
[00:11:34] Paul Roetzer: Like it’s, it is an inevitable outcome and that’ll likely start happening at some time in
[00:11:38] Paul Roetzer: 2025. We’ve already heard the story of Klarna, you know, their AI agent, basically. Customer service agent doing the work of 700 reps. Like that, that’s just straight up. They need fewer people doing the work. So that’s kind of like how we set the stage timeline wise.
[00:11:54] Paul Roetzer: So I think what, what we’re seeing with HubSpot others
[00:11:59] Paul Roetzer: is this [00:12:00] very early version of these very specifically tuned agents designed to do very specific tasks. things. They’re not generally capable of doing many things.
[00:12:08] Paul Roetzer: They’re trained to be a sales service rep bot or a agent bot or a coding bot or things like that, or service rep bot.
[00:12:16] Paul Roetzer: so they’re trained to do these things. But now let’s go back to open AI’s levels of AI. We talked about this on recent podcasts. So level one is chatbots, which is
[00:12:24] Paul Roetzer: what we’ve had with ChatGPT. Level two is reasoners, what In theory, O1 gives us their new model that does reasoning. And we talked about last episode, it’s a very early version of it.
[00:12:35] Paul Roetzer: Like we’re at like the GPT stage of reasoning capability, but it’s probably going to accelerate pretty fast.
[00:12:42] Paul Roetzer: Level three for open AI is agents, systems that can take actions. So, Then level four is
[00:12:47] Paul Roetzer: innovators and then level five is organization. So we’re, we’re very much in this level two to level three range for what OpenAI considers the levels of AI.
[00:12:57] Paul Roetzer: related to OpenAI, so we go [00:13:00] back to episode 83 of the podcast, February 13th of this year, and we recapped their World of Bits paper, so back in 2017. So OpenAI has been working on agents since their foundation, since forming, in 2015, 16.
[00:13:16] Paul Roetzer: So in that paper, Andrej Karpathy, whose name you hear a lot on the podcast, led a team that looked into this building these agents that could do things that could take actions like booking flights
[00:13:27] Paul Roetzer: and completing forms, by simulating the use of a keyboard and mouse, where it could go in and do all these general tasks.
[00:13:33] Paul Roetzer: And their conclusion at the time was it just wasn’t ready yet. But then in 2022, he did an interview with Lex Fridman and said, we think we
[00:13:41] Paul Roetzer: might be there.We think that language was the unlock by building, models, AI models that could understand and generate human language. We think we now can teach these agents how
[00:13:51] Paul Roetzer: to do things. And so that, that kind of led me to. To that. so Sunday, September 22nd,
[00:13:57] Paul Roetzer: listening to a podcast with [00:14:00] Brett Taylor, on the NoPriors podcast. And so, if you don’t know who Brett Taylor is, he is the co founder of Sierra, Sierra Works, I think is the name of the company. but more importantly, he is, the, co, he was the co CEO of Salesforce prior to this.
[00:14:17] Paul Roetzer: And prior to Salesforce, he had founded
[00:14:19] Paul Roetzer: Quip and he was the chief technology officer for Facebook, and he spent the early parts of his career at Google, where he co created Google Maps. but
[00:14:28] Paul Roetzer: maybe the most important, he is on the OpenAI board. He is the surviving member of the changeover of the OpenAI board.
[00:14:36] Paul Roetzer: His co-founder at Sierra is, Clay Baver, who we’ve never talked about on the podcast. We’re actually haven’t followed Clay prior to this. but
[00:14:44] Paul Roetzer: he was at Google for 18 years, recently led Google Labs. He started and led their Google AI VR, AR VR effort, Project Starline and Google Lens, interestingly enough.
[00:14:58] Paul Roetzer: and he led the product and [00:15:00] design teams for Google Workspace. So the, all this is to say is Clay and Brett. are
[00:15:05] Paul Roetzer: highly, highly connected to what is going on
[00:15:08] Paul Roetzer: in this world and where these models can go. And so interestingly enough, I had flagged this episode to listen to, not knowing we were like It was about agents and that we were going to be talking about agents the next day, but this episode with Brett, this interview literally starts
[00:15:25] Paul Roetzer: off with do agents work today? So
[00:15:27] Paul Roetzer: it just, they dive right in because his company, Sierra, builds company agents. So I thought, Brett, who again is on the board at OpenAI, building a agent company on top of OpenAI’s technology. Has probably more insight than anyone outside of Sam
[00:15:42] Paul Roetzer: Altman into the strategy of what they’re building and where it’s going. I think he’s the chairman of the board. We could, maybe you could double check that, Mike,
[00:15:49] Paul Roetzer: but I’m pretty sure Brett is the main guy. So they ask him straight up, like, do agents work today?
[00:15:56] Paul Roetzer: And how do you define them? And so he, again, [00:16:00] Again, this is the big sticking
[00:16:01] Paul Roetzer: point to me right now is like, what are we talking about when we talk about agents? So he said
[00:16:05] Paul Roetzer: the agents mean something different in academia than I think they mean in industry right now.
[00:16:10] Paul Roetzer: I think both definitions are important. Just starting what I view as the classic academic
[00:16:15] Paul Roetzer: definition is an agentic system. is one where software can reason and take action autonomously. And it comes from the word agency. So agency, again, if people
[00:16:26] Paul Roetzer: aren’t familiar with kind of the technical side of this, is like, if we think about agency in humans, our agency is our
[00:16:32] Paul Roetzer: ability to act independently, to make decisions, to influence the environment with, with it, with, with, within
[00:16:39] Paul Roetzer: which we live. So we have control of our choices and and we have responsibility for the outcomes of those choices. So the concern in AI, or in some cases, the pursuit in AI is to give the AI agency, to give the AI, the ability to make its own choices, to influence its outcomes, [00:17:00] to influence its
[00:17:00] Paul Roetzer: environment. And that is the idea of.
[00:17:03] Paul Roetzer: of AI agents is to give the AI agency, to give it the ability to take actions and have responsibility for the outcomes of those actions.
[00:17:13] Paul Roetzer: So, he then talks about there being three sort of categories of agents that, that we’re working on or that are kind of like, in the world today. And this, this then leads back to what exactly does Salesforce have? What does HubSpot have? What’s the connection there? So we have personal agents, which is kind of what we talk about
[00:17:32] Paul Roetzer: with Apple intelligence on your phone.
[00:17:34] Paul Roetzer: This kind of general agent that can go and do a bunch of things, within your environment, within your phone. Now that may
[00:17:42] Paul Roetzer: end up being like 50 agents working in symphony together. That we don’t see, we just see the one main agent,
[00:17:50] Paul Roetzer: but there may be all these other agents working together behind the scenes. So personal agents are kind of more general
[00:17:57] Paul Roetzer: intelligence. It’s what, what Google’s probably working [00:18:00] on with some of their bigger initiatives. Um, Maybe Microsoft, certainly OpenAI is what they’ve been pursuing
[00:18:06] Paul Roetzer: all these years. Then there’s persona based agents, which are kind of trained to do a specific thing.
[00:18:11] Paul Roetzer: So like Harvey, the law, the company that’s building a legal AI agent, they’re trying to do like these persona based agents. And then there’s company agents that sort of represent your, brand. your brand.
[00:18:23] Paul Roetzer: And so Brett, I really recommend people go listen to the whole podcast. Brett does a great job of just breaking down where we’re at.
[00:18:30] Paul Roetzer: But what he basically said is with personal agents, we’re kind of like earliest
[00:18:34] Paul Roetzer: like this idea of building these general agents that can go and, like, let’s say I have
[00:18:38] Paul Roetzer: a single general agent living within Google, we’re, you know, Google workspace customers. If I had a single agent, that learned me, learned to manage my inbox, learned to handle my schedule, learned to do all these things.
[00:18:53] Paul Roetzer: That would be like a personal general agent that I entrusted to like learn everything about me, have [00:19:00] access to all of my data, all of my systems. That’s what he’s saying is the earliest. The people who are going to build that are likely going
[00:19:08] Paul Roetzer: to be Google, OpenAI, Microsoft. I don’t, I don’t even know how, I can’t imagine like Salesforce doesn’t have access to enough data.
[00:19:17] Paul Roetzer: Like you, you need a cloud platform that has access to basically everything. Apple, you know, those are the companies that are
[00:19:22] Paul Roetzer: going to build those. The persona based agents is probably more of what you’re going to interact with,
[00:19:30] Paul Roetzer: more of what we’re seeing Salesforce and HubSpot building. And that is agents to do specific things.
[00:19:36] Paul Roetzer: These are easier to emerge sooner. Because they’re confined in their goals and the data they need to have
[00:19:44] Paul Roetzer: access to. So if I want to do an email agent,
[00:19:47] Paul Roetzer: that’s going to help me develop and send emails in HubSpot. I maybe build or go to agent. ai and
[00:19:53] Paul Roetzer: I find one, or I just go right within HubSpot. It doesn’t have access to my personal Gmail calendar.
[00:19:59] Paul Roetzer: [00:20:00] Or Google calendar. doesn’t have
[00:20:01] Paul Roetzer: access to my Gmail. It doesn’t have access to anything else. It just has access to the email capabilities it needs to help me do email. Customer service, PersonAgent, same deal. It has access to CRM data, has
[00:20:13] Paul Roetzer: access to call transcripts, whatever it needs to do. Customer support.
[00:20:19] Paul Roetzer: So he talks a lot about that and then he gets into, this idea of these company agents. And he literally says in
[00:20:25] Paul Roetzer: 2025, existing digitally will probably mean
[00:20:28] Paul Roetzer: having a branded AI agent that your customers can interact with to do everything that they can do with your website. You know, talk about your products and services, do e commerce, do customer service.
[00:20:38] Paul Roetzer: So, that’s what his company, Sierra, is building on top of, I assume, OpenAI’s technology. And not just
[00:20:44] Paul Roetzer: the tech we see today, but he’s probably has access to the tech that the rest of us will see tomorrow. And he’s building these capabilities on top of that. He gets into, like, retrieval augmented generation and how that’s not enough, how, like, actually you have to take a goals and [00:21:00] guardrails approach, which I thought was kind of a fascinating way to understand it.
[00:21:03] Paul Roetzer: So in, in his world, those are the three ways he thinks about agents. But at the end of the day, I think what we kind of have is what we’ve always had with
[00:21:13] Paul Roetzer: AI. You have narrow intelligence, narrow agents. And then you have. General Intelligence and General Agents. What we will
[00:21:20] Paul Roetzer: all interact with over the next one to two years is far more likely narrow. They’re going to be the persona based agents trained to do specific things. That’s what you’re going to build and use in Agent.
[00:21:31] Paul Roetzer: AI and HubSpot. It’s what you’re going to build and use in Salesforce. It’s what you’re going to build and use in Microsoft, who
[00:21:36] Paul Roetzer: had an announcement last week, about their updates to Copilot and changes they’re making there.
[00:21:43] Paul Roetzer: They said, let’s see, Co Pilot Agents and Microsoft 365 Co Pilot. We’ll put a link to this. Microsoft said, whether you’re a user, developer, IT professional, Microsoft
[00:21:52] Paul Roetzer: Co Pilot Studio offers a comprehensive platform for creating, managing,
[00:21:56] Paul Roetzer: and deploying Co Pilot Agents. The two new things [00:22:00] they announced was Co Pilot Agents, built
[00:22:02] Paul Roetzer: in Copilot Studio can now be published directly into Microsoft 365 Copilot. And you can share them with your coworkers,
[00:22:08] Paul Roetzer: I assume. And they’re unveiling Copilot Studio experience that empowers users of any skill level to create Copilot agents.
[00:22:15] Paul Roetzer: within SharePoint and Microsoft, And the examples they gave was an IT help desk agent, a device refresh agent, a lead gen agent, a project tracker agent.
[00:22:24] Paul Roetzer: you get the point. Whatever you do, task wise, they’re going to build agents for it.
[00:22:29] Paul Roetzer: Now, if we rewind back to episode 92, April 16th of this year, that was,
[00:22:34] Paul Roetzer: Mike, when we talked about Google Cloud Next Conference, Sundar on stage talking about what AI agents. So they define them as intelligent
[00:22:44] Paul Roetzer: entities that take actions to achieve goals.
[00:22:47] Paul Roetzer: And they specifically talked about the idea of agents connecting with other agents. They introduced Vertex AI, Vertex AI Agent Builder. They talked about them as employee agents that, again, [00:23:00] We’re going to say they’re going to be your
[00:23:01] Paul Roetzer: coworkers. In many cases, they may be. You may build an org chart that actually infuses
[00:23:07] Paul Roetzer: AI agents as part of that org chart. You may also replace people
[00:23:11] Paul Roetzer: with these things. It’s kind of like the unspoken word. at the time they talked about, they showed examples of customer agents, employee agents, creative agents. Data agents, code agents, and security agents. Those were their broad six categories.
[00:23:25] Paul Roetzer: and then you would like, branch off from within there. So the point of all of this is you are going to
[00:23:32] Paul Roetzer: be hearing a ton about AI agents. We are going to enter this phase where everyone’s going to have slightly different definitions of what they are. Um, they’re going to talk about them as.
[00:23:44] Paul Roetzer: agents. Additions to, augmentations to what humans are capable, which is not lying.
[00:23:50] Paul Roetzer: They are, but by this
[00:23:53] Paul Roetzer: time next year, I, I can almost promise you, we are going to have podcast episodes talking about agents as [00:24:00] replacements. to workers. Like it is an
[00:24:02] Paul Roetzer: inevitable outcome. So they’re going to be awesome.
[00:24:06] Paul Roetzer: They’re going to make our lives more efficient, more productive. You’re going
[00:24:09] Paul Roetzer: to love using these things. They’re going to start getting really reliable next year, especially the more narrow they are, the better you train them, the more data they have.
[00:24:18] Paul Roetzer: But I will say and kind of wrap up here, Mike, in case if you have any thoughts, I think while
[00:24:24] Paul Roetzer: I’m highly confident they will replace people. I’m also highly confident people will have new or evolved jobs training and managing these agents. Right.
[00:24:35] Paul Roetzer: Because who better to build them and to, to know how they’re performing and to improve them than the people who have domain expertise and knowledge in that space. And there’s a chance
[00:24:45] Paul Roetzer: that maybe that offsets it. Like maybe agents become such a massive thing within organizations. And it just all, like, maybe it just levels up productivity so much in companies that it offsets job loss.
[00:24:58] Paul Roetzer: I don’t know. Like [00:25:00] maybe.
[00:25:00] Paul Roetzer: Maybe that happens, and I hope it does, but the reality is we’ll need fewer people doing existing work if we
[00:25:07] Paul Roetzer: can create enough new work and increase productivity enough with these things, then by this time next year, people listening to the show,
[00:25:14] Paul Roetzer: like, there may be people, some of you who are AI agent managers and AI agent builders, and you know, if I was
[00:25:20] Paul Roetzer: running an agency right now, I would be racing to figure out how to build AI agents for
[00:25:25] Paul Roetzer: clients, because I think that’s, there’s an enormous opportunity for the next. Who knows five, 10 years to, to be playing in this space.
[00:25:33] Paul Roetzer: So it’s, it’s, I get that that’s probably a lot of information,
[00:25:37] Paul Roetzer: but I think it’s really critical that people have this foundation of what are we talking about when we’re talking about agents?
[00:25:42] Paul Roetzer: Where are we actually with them? And why is it the only thing all these,
[00:25:47] Paul Roetzer: tech companies are talking about right now?
[00:25:49] Mike Kaput: Yeah, and that’s exactly why I wanted to have this conversation is really to identify first and it sounds like the answer is yes, can
[00:25:58] Mike Kaput: agents, at least in some [00:26:00] areas, do certain tasks autonomously? And we are increasingly getting to a world where that’s the case. Now,
[00:26:06] Mike Kaput: Like you mentioned, there’s lots of nuance, there’s
[00:26:09] Mike Kaput: lots of reliability issues, But if that world comes to pass, comes back to that idea we always talk about. A job is a bundle of tasks, right? So you different tasks, suddenly
[00:26:20] Mike Kaput: it’s not just you using AI to do the task. Agents may do the task. very well do many of
[00:26:26] Mike Kaput: those tasks, which is going to extrapolate those effects however you will. I mean, that will be disruptive.
[00:26:32] Paul Roetzer: Yes, human, human built agents doing human tasks with human oversight is kind of where we are in very specific areas and domains. And that’s, that’s kind of what’s being built today.
[00:26:44] Mike Kaput: And I mean, and you know, this could be a whole other podcast episode, but it’d be worth maybe in a future episode us
[00:26:49] Mike Kaput: exploring what does that mean for hiring? Because the skill set of running A team of agents we’re building them is very different than like being a great copywriter, for [00:27:00] instance, or something like that.
[00:27:01] Paul Roetzer: Yeah. And you’re not going to be highly technical, need to be highly technical to build these things. So like
[00:27:05] Paul Roetzer: right now I haven’t obviously gone into AgentForce, which I don’t even know if AgentForce Is a platform or like, I don’t, I still don’t know what exactly AgentForce is.
[00:27:14] Paul Roetzer: I think it’s just more of a brand overarching, but anyway, like I haven’t tried to go in and build one there.
[00:27:19] Paul Roetzer: Vertex AI Agent Studio, I did back when it was first introduced. And while they say it’s no code, you, you need technical. Capabilities
[00:27:28] Paul Roetzer: to to play in there, but I do think that in the next year or so, we’re largely going to get to a no code environment, kind of like we’ve talked about with our custom
[00:27:38] Paul Roetzer: GPTs, Mike, like I, I built three custom GPTs with zero knowledge of how to code. I envisioned that being more realistic. So I think that while today, most.
[00:27:48] Paul Roetzer: agents, quote unquote, that are built are built by people with a programming background. I could see within
[00:27:54] Paul Roetzer: 12 months, the vast majority of agents being built are by the average, knowledge worker with no coding [00:28:00] ability. Like, I really think we are racing toward a very low code, no code future where most instances is someone with domain expertise who understands the process and the steps that you have to
[00:28:11] Paul Roetzer: go through that just builds an agent to assist them in doing the work or to replace, you know, 50, 80, 90 percent of the tasks that are involved in doing something.
[00:28:21] Mike Kaput: That aspect of it is really exciting to me, despite the uncertainty, it’s like, I think it’s a good reminder, regardless of how smart the engineers are out there. For you non engineers, that
[00:28:31] Mike Kaput: doesn’t mean they know anything about how to do your job. You have a better idea already using your domain expertise about what product or agent or system to build to
[00:28:43] Mike Kaput: achieve what you do or help you with what you do better than anybody else. The thing we’ve just lacked is the ability. to do it until now,
[00:28:50] Paul Roetzer: which you and I have talked about is like a really exciting thing. So like the ability
[00:28:54] Paul Roetzer: for me to sit down and build problems, GPT as an example, without having to call up a developer [00:29:00] or find somebody that I
[00:29:00] Paul Roetzer: trusted to do it without having to go through and do all the technical sides of the data. that was very empowering.
[00:29:06] Paul Roetzer: Like that, that was a wonderful feeling to just take some knowledge. And to put it into a tool. And so I do think like, while, while
[00:29:14] Paul Roetzer: I am, you know, I talk about the impact on jobs, I do believe there’s this like incredible side to this, that it’s going to empower people to build things that they just couldn’t have done before.
[00:29:25] Paul Roetzer: And that’s going to be awesome for a lot of people.
[00:29:28] Runway
[00:29:28] Mike Kaput: All right, so on to our third big topic. Remember HubSpot slash Salesforce kind of teeing up topic one, AI agents in
[00:29:35] Mike Kaput: the enterprise topic two. Third is taking a bit of a left turn here because we also have some big news coming out of the creative world.
[00:29:45] Mike Kaput: AI video generation company Runway has just inked a groundbreaking deal with Lionsgate, which is a major Hollywood studio behind popular movie franchises like John Wick and Twilight.
[00:29:57] Mike Kaput: Now what’s notable here is this partnership [00:30:00] marks the first public collaboration formally between a generative AI startup and
[00:30:04] Mike Kaput: a major Hollywood studio, though there have been plenty of conversations between studios and OpenAI, etc.
[00:30:11] Mike Kaput: Now under this agreement, Runway will train a custom video model using Liongate’s extensive movie catalog.
[00:30:20] Mike Kaput: The studio’s creative talent, including
[00:30:21] Mike Kaput: filmmakers and directors, will gain access to this AI model to enhance their work.
[00:30:27] Mike Kaput: Vice Chairman at Lionsgate, Michael Burns, emphasized the potential for this technology to augment
[00:30:34] Mike Kaput: the creative process. Now, Runway
[00:30:35] Mike Kaput: is also exploring ways to license these models as templates, allowing individual creators to build and train
[00:30:42] Mike Kaput: their own custom models. Now, Paul,we’ve talked about this issue a couple different times in a few different contexts, but this
[00:30:49] Mike Kaput: seems notable given that it’s the first major Hollywood studio to just kind of openly sign this kind of deal.
[00:30:55] Mike Kaput: Like, does this open some floodgates? Like, is it now acceptable to embrace [00:31:00] AI in Hollywood creative work, at least?
[00:31:01] Paul Roetzer: Yeah. And the Wall Street Journal article on this actually went deep. We’ll put the link in
[00:31:07] Paul Roetzer: and they basically talked about how Lionsgate felt they were, they had to do this, that this is probably one of many near term deals that’s going to be announced, that OpenAI, Both Sora has been in talks with major studios that, yeah, I would imagine Google’s probably in talks with different studios with VEO, I think is their, video generation model. sue
[00:31:46] Paul Roetzer: And so I think on images, on video, on music, one by one, at the end of the day, these companies
[00:31:53] Paul Roetzer: have two choices, sue, these AI model companies for training on their data or work with them [00:32:00] and make money. And I think one by
[00:32:02] Paul Roetzer: one, they’re They’re going to choose to work with them and make money model. I think that this is, there’s going to be a lot
[00:32:10] Paul Roetzer: more of these. It’s going to be interesting to see how it plays out. Like the idea of Runway being able to train on Lionsgate data and then potentially, Maybe through a joint venture, offer that custom train model
[00:32:21] Paul Roetzer: to other Movie studios who can’t do the big deals. Like, there’s a lot of layers to this that
[00:32:26] Paul Roetzer: I find really intriguing. And that’s why I wanted to, you know, throw this in as a main topic is I think it.
[00:32:32] Paul Roetzer: is a prelude to a lot of other conversations we’re going to be having over the next year, like on a different episodes about the next big deal that was done.
[00:32:41] Paul Roetzer: and then. As we start to learn how these deals play out and the implications of them. Cause even in the wall street journal article, they said that, there was initially, they’re going to use Lionsgate’s going to use it for internal purposes, like
[00:32:54] Paul Roetzer: storyboarding, laying out
[00:32:55] Paul Roetzer: a series of graphics to show how the story unfolds and eventually creating backgrounds and special [00:33:00] effects like explosions for the big
[00:33:01] Paul Roetzer: screen. So they’re taking this kind of long game approach here of let’s get this model trained and
[00:33:06] Paul Roetzer: then we can start using it ourselves. If it works really well, we’ll start putting it into the films and maybe we’ll create background characters with it. we’ll create background scenes, extend scenes, like after production has ended and we miss something, we can go back and recreate it.
[00:33:21] Paul Roetzer: Like, there’s so many applications of this in movie production and video game production is the other one we’ll
[00:33:26] Paul Roetzer: be talking a ton about. So this is going to be a big topic moving forward and it sort of runs in parallel with the legal challenges and copyright stuff. But I think
[00:33:37] Paul Roetzer: you’re going to get a lot more headlines about this sort of thing in the near future than you are any progress on the legal fronts that impacts the training of these models.
[00:33:46] Mike Kaput: Yeah. and what I like about this topic too is
[00:33:49] Mike Kaput: Regardless of how you feel about it, which isn’t very valid and important if you think it’s good or bad or ugly, it shows you what’s actually happening.
[00:33:58] Mike Kaput: Like, you look at what [00:34:00] people are doing, not what they’re necessarily saying. Like,
[00:34:03] Mike Kaput: this is clearly we can’t beat them, so we’re joining them. Move. They say it’s going to augment everything in creative work required
[00:34:12] Mike Kaput: for these, but also recall, I’m, I’m no expert on Hollywood economics, but I know they’re not great. A
[00:34:18] Mike Kaput: lot of times, you know, they’re under immense pressure, so when you get these factors coming together, I don’t know, you do the math.
[00:34:25] Paul Roetzer: Yeah, and they talked in the Wall Street Journal article about it being able to save like tens of millions of dollars a year in production costs and different things. So yeah, there’s a real business case for this.
[00:34:36] BlackRock AI Infrastructure
[00:34:36] Mike Kaput: All right, let’s dive into some rapid fire. First up, we have some news that BlackRock, which is the world’s largest
[00:34:43] Mike Kaput: asset manager, is partnering with Microsoft to launch a 30 Billion fund dollar fund aimed at investing in AI infrastructure. So this fund is focusing on building data centers and energy projects, and it’s designed to do something
[00:34:58] Mike Kaput: we’ve talked about quite a bit on the [00:35:00] podcast, meet the growing energy and computational demands. of AI. This fund is particularly notable for its scale. It represents one
[00:35:09] Mike Kaput: of the biggest investment vehicles ever raised on Wall Street, according to the Financial Times.
[00:35:14] Mike Kaput: In addition to Microsoft, NVIDIA, and MGX, a financial company backed by Abu Dhabi, are also general partners in this fund. NVIDIA will provide technical expertise in chipmaking. MGX is
[00:35:27] Mike Kaput: going to offer financial support. And this basically comes as AI technologies are increasingly straining existing energy and data infrastructure. The
[00:35:38] Mike Kaput: International Energy Agency predicts electricity consumption by data centers could more than double by 2026. And this is why we’re also hearing at the same time, another really interesting
[00:35:49] Mike Kaput: report, which is that a company called Constellation Energy. Has announced plans to restart the Unit 1 reactor at the Three Mile Island Nuclear Plant, and they are [00:36:00] selling its energy output to Microsoft. This agreement
[00:36:03] Mike Kaput: between Constellation and Microsoft is one of the largest of its kind. Microsoft is securing a 20 year contract to power its data centers with carbon.
[00:36:13] Mike Kaput: Free energy. Now, data centers are expected to consume 8 percent of the US’s total electricity demand by 2030, up from
[00:36:23] Mike Kaput: 3 percent today, according to one forecast from Goldman Sachs. So, Paul, these are, in a few ways, some pretty incredible numbers.
[00:36:32] Mike Kaput: Like we’ve talked about the importance of physical infrastructure to support AI’s progress.
[00:36:37] Mike Kaput: Like, how big a deal is this particular investment initiative?
[00:36:42] Paul Roetzer: Yeah, so we’ve talked a number of times starting back on episode 87 about this idea that, there’s accelerants to AI advancement. And one of them that we highlighted then, we’ve talked
[00:36:53] Paul Roetzer: many times about since, is massive investments in energy and infrastructure. You know, we become numb to [00:37:00] how much a billion is, like
[00:37:01] Paul Roetzer: I said on a recent episode that the 10 billion chip act was like, Great, but it’s, it’s nothing like, you know, half
[00:37:09] Paul Roetzer: jokingly, there was word that Sam Altman was trying to raise 7 trillion. Like, that’s why I said like, you know, OpenAI raising
[00:37:17] Paul Roetzer: six and a half billion at 150 billion valuation, like, it sounds like a ton of money and
[00:37:23] Paul Roetzer: it is. But relative to the money that’s going to be spent over the next decade to put the US economy, and not just the US, like, you know, globally, to put
[00:37:33] Paul Roetzer: us in a position to build out the infrastructure that these AI companies believe is needed.
[00:37:40] Paul Roetzer: For the intelligence explosion that we are entering, these numbers are crazy. And a hundred billion isn’t even going to seem like a lot in three years. Like, you’re going to hear about a trillion dollar investment in infrastructure within the next like two to three years. Like, it’s going to happen. I don’t know, again, I don’t know if the U.
[00:37:59] Paul Roetzer: government [00:38:00] has the will to, to do it themselves. I kind of worry that it’s like private,all private at this point. I feel like the government
[00:38:09] Paul Roetzer: needs to do something bigger than it’s doing. But yeah, I mean, we’re going to hear numbers like
[00:38:13] Paul Roetzer: 100 billion getting thrown around, like it’s no big deal. And it is a huge deal. but yeah, these are, these are big numbers and,
[00:38:23] Paul Roetzer: and we’re going to be seeing a lot. More of it coming from the major players in the space because they, they need
[00:38:30] Paul Roetzer: The energy infrastructure to build the data centers
[00:38:33] Paul Roetzer: that, that will power not only the training of the future models, but the inference when all of the devices in our lives have this intelligence on demand and it’s, it’s coming fast.
[00:38:45] Sam Altman Goal 3
[00:38:45] Mike Kaput: All right. so in our next topic, we have said time and time again, you have to take even offhand posts. Online from Sam Altman. Seriously,
[00:38:54] Mike Kaput: You know, in the past, we’ve talked about he’s teased some major AI advancements like Strawberry slash [00:39:00] O1 with
[00:39:00] Mike Kaput: sometimes cryptic or tongue in cheek posts on X, and I think he might’ve just dropped another one. On September
[00:39:07] Mike Kaput: 17th, he posted on X, quote, incredible outperformance on goal three, even though it took a while. And then he linked to a post
[00:39:15] Mike Kaput: from OpenAI written way back in 2016 about the company’s technical goals. This post outlines four technical goals the company aimed to achieve, including
[00:39:26] Mike Kaput: measuring their progress, building a household robot, and solving a wide variety of games using a single agent.
[00:39:33] Mike Kaput: Now, one of the goals that I did not mention is goal three, that Altman is referencing. This goal is, quote, build an agent with useful natural language understanding. In
[00:39:45] Mike Kaput: the post, it’s described as, quote, we plan to build an agent that can perform a complex task specified by language and ask for clarification about the task if it’s
[00:39:55] Mike Kaput: ambiguous. Altman didn’t really offer a ton more context to this
[00:39:59] Mike Kaput: [00:40:00] post, but interestingly, the 2016 article itself is like, definitely worth a little read, it’s like opening up a bit of a time capsule, because Altman is an author, so is Greg Brockman. And so is Ilya Sutskever and Elon
[00:40:14] Mike Kaput: Musk. So the original kind of OGs of OpenAI long before OpenAI was the size and scale it is today.
[00:40:23] Mike Kaput: So Paul, can you kind of maybe give us a sense of like what’s going on here? Back to the agent conversation, like, is Sam saying that this stuff already exists? Like, in the models we can access, like, O1
[00:40:35] Mike Kaput: can kind of do this, though it’s not really an agent? Or is he like teasing something that’s coming next?
[00:40:41] Paul Roetzer: Well, I, I think it’s probably a combination of, again, I think their next model is probably. Done and in testing red teaming. and it may be alluding to kind of what they’re going to be bringing out. Cause again, we haven’t seen the full [00:41:00] O1 model. We’ve only seen preview and mini. And so
[00:41:03] Paul Roetzer: it’s very possible He’s just sort of touting that. Yeah, they’ve made a ton of progress here. I think it’s interesting to look at this list. Like they’re not even pursuing goal two and goal four anymore and i don’t
[00:41:13] Paul Roetzer: I think goal one has become that levels of five. artificial intelligence, the levels that we talked about earlier. the household robot obviously has become, you know, Elon Musk has sort of picked that
[00:41:24] Paul Roetzer: up and run with that with Optimus and others are trying to pursue that and there’s been
[00:41:29] Paul Roetzer: talks actually, Andrej Karpathy talked about the robot thing on a recent podcast and how they were pursuing that early on and they, they moved on
[00:41:36] Paul Roetzer: And then the, you know, the
[00:41:38] Paul Roetzer: games because of DeepMind’s success with AlphaGo, they thought that was a path they were going to pursue.
[00:41:44] Paul Roetzer: Yeah, I would say that they have, sort of singularly focused on Goal three,
[00:41:48] Paul Roetzer: you know, since probably the Transformer invention in 2017 is when they really sort of shifted their focus on, into language.
[00:41:56] Paul Roetzer: yeah, I, I mean, Sam is always[00:42:00] teasing something that’s coming. He rarely takes a victory lap for things that they’re, they’ve just done.
[00:42:07] Paul Roetzer: And so I think it’s probably more alluding to what we’ll see
[00:42:10] Paul Roetzer: in the near future than what they’ve previously showed us. But who knows with Sam, sometimes he just tweets things just to
[00:42:19] Mike Kaput: Yeah, true.
[00:42:20] Paul Roetzer: almost always an alluding to something that’s coming.
[00:42:24] YouTube + AI
[00:42:24] Mike Kaput: So in our next topic today, YouTube’s latest Made on YouTube event unveiled a series
[00:42:30] Mike Kaput: of AI powered features to empower creators, and a key highlight of these updates is the integration of Google DeepMind’s VO
[00:42:38] Mike Kaput: video generation model into YouTube Shorts through a feature called DreamScreen. Now, DreamScreen is an existing capability. It was released last year. It basically allows creators to generate
[00:42:50] Mike Kaput: backgrounds in their Shorts videos. But now, VEO, which is Google DeepMind’s most capable model for generating video, it will [00:43:00] now be integrated into YouTube Shorts later this year.
[00:43:03] Mike Kaput: Their advanced image generation model, ImageN3, will also be integrated into the DreamScreen feature. So from Google DeepMind, they said, quote, DreamScreen uses
[00:43:13] Mike Kaput: ImageN3 to generate four distinct images based on a text prompt, offering creators various options for style, composition, or aesthetic. After selecting an image, Veo will turn it into a high quality video background.
[00:43:27] Mike Kaput: This functionality designed to inspire creativity will become even more powerful in 2025 when users will
[00:43:33] Mike Kaput: be able to generate short, standalone video clips, such as
[00:43:37] Mike Kaput: a cinematic scene, directly from their imagination. The AI generated content will also be labeled with Google DeepMind’s SynthID watermarking technology.
[00:43:49] Mike Kaput: Another exciting development, they now have an inspiration tab that’s revamped with AI powered brainstorming tools to help creators generate new video
[00:43:57] Mike Kaput: ideas, titles, thumbnails, and [00:44:00] outlines. So Paul, we talk a lot about Google DeepMind on the pod. It’s interesting and
[00:44:05] Mike Kaput: welcome to see its models being very kind of clearly commercialized within YouTube. I think what stood out to me is that we’re easing into you’re going to be able to generate videos. on YouTube with AI. Like, are we,
[00:44:19] Mike Kaput: Should we expect to see that become much more prevalent on YouTube?
[00:44:22] Paul Roetzer: Yeah, it’s going to be an interesting shift. I do feel like we’re just racing toward this future where it has generated stuff, it’s just going to be everywhere.
[00:44:31] Paul Roetzer: I think the interesting play here is like we talked last week, and we’ll touch on Notebook LLM again from Google, is when they take these models, and it’s nice to have this general,
[00:44:42] Paul Roetzer: like going to ChatGPT and you can do anything you want, or going to Google Gemini and
[00:44:45] Paul Roetzer: you can create anything you want. But when you successfully
[00:44:48] Paul Roetzer: integrate this technology into existing workflows, that just becomes a natural extension of how people interact with
[00:44:55] Paul Roetzer: a tool or a platform or a social media channel. Adoption just [00:45:00] Skyrockets. And so that’s
[00:45:01] Paul Roetzer: where you always go back to, like, well, who has the distribution? It’s Google and Meta and Apple that, you know, control the devices and the software we use every day. And then going back to
[00:45:11] Paul Roetzer: the agent conversation, where do you interact with it most? Well, you’re already in HubSpot, you’re already in Salesforce.
[00:45:17] Paul Roetzer: And so that, I, I’m fascinated on that side of it too, just how this
[00:45:21] Paul Roetzer: distribution and adoption plays out, um, to where, you know, one year out, like everyone on Facebook and, you know, YouTube and Instagram and WhatsApp, like, you can’t not use AI. It’s going to be so
[00:45:36] Paul Roetzer: infused in everything we do. It’s just going to be, it’s going to be everywhere. So, yeah.
[00:45:41] Mike Kaput: And again, look at the incentives, right? It’s like, whether or not there’s outcry about eventual AI generated videos
[00:45:48] Mike Kaput: on YouTube, creators on YouTube that rely on it are not going to stop using it. So
[00:45:53] Paul Roetzer: For sure. Yeah, it’s, it’s going to be fascinating. And I’m with you on the,
[00:45:59] Paul Roetzer: they’re [00:46:00] giving a timestamp now on the integration of video generation. And so you do wonder if either a fall event this year or a
[00:46:06] Paul Roetzer: spring event next year, like the Google Cloud Next in March, April, whenever they do it next year, if that
[00:46:12] Paul Roetzer: isn’t when we get, Our, our
[00:46:14] Paul Roetzer: video capabilities, I would assume sooner, because I can’t imagine OpenAI is gonna wait a full year with SOA before they release
[00:46:20] Paul Roetzer: that. I, OpenAI has a dev day coming up on October 1st, and I wonder if we won’t
[00:46:25] Paul Roetzer: maybe see what’s coming this fall from OpenAI at that, that dev day.
[00:46:31] LinkedIn Training on User Data
[00:46:31] Mike Kaput: All right. so next up, LinkedIn has come under some scrutiny for using user data
[00:46:37] Mike Kaput: to enhance its generative AI products without first updating its terms of service. This was revealed in a
[00:46:43] Mike Kaput: recent report from 404 Media. This past Wednesday, several LinkedIn users observed a setting indicating that their data was being used for AI improvements, even though the company’s terms did not reflect this practice. After 404 reported on this, LinkedIn
[00:46:59] Mike Kaput: [00:47:00] confirmed. that they were making an update to their terms shortly at the time the report
[00:47:04] Mike Kaput: broke, they now appear to have done so. The Verge reports the updated policy now states that LinkedIn may use personal data to train AI models, develop services, personalize user
[00:47:15] Mike Kaput: experiences. To stop LinkedIn from using your data for AI model training in the future, you have to go to your account settings under the
[00:47:23] Mike Kaput: data privacy tab and turn off quote data for generative AI improvement. as an option. However,
[00:47:30] Mike Kaput: this action only prevents future data usage. It doesn’t reverse any data processing that has already occurred. For users in the EU, LinkedIn states that their
[00:47:40] Mike Kaput: data is not included in the AI training sets, given the stricter regulations in those areas. Now, Paul, I wish I could say this surprised me, but it doesn’t, since I feel like we’ve
[00:47:51] Mike Kaput: seen it happen time and time again with different platforms. Like, you’re a big LinkedIn user. Like, how did you feel reading this?
[00:47:58] Paul Roetzer: Yeah, I am a little surprised it’s LinkedIn. Um. You know, I think that there’s some brands you just assume you can trust a little bit more, but I,
[00:48:07] Paul Roetzer: by this point, I think we’ve all learned that if you put something publicly up, I just assume somebody’s training on it. If it’s not LinkedIn, someone’s scraping
[00:48:14] Paul Roetzer: LinkedIn data to train on it.
[00:48:15] Paul Roetzer: So I you know, I, honestly, there’s a part of me that’s like, Oh, we didn’t know that already. Like that they were training on our data. I kind of assumed it. You can go to the Responsible AI Principles. LinkedIn actually has on their blog from February, 2023, which is right before this. You know, right after ChatGPT came out, shared
[00:48:33] Paul Roetzer: LinkedIn responsibility principles, and they, you know, say that uphold trust, promote fairness and inclusion, provide transparency, oops, embrace accountability.
[00:48:42] Paul Roetzer: Like, so, you know, again, it’s a reminder too that
[00:48:45] Paul Roetzer: responsibility principles are only as good as, you know, people actually understanding, them and following them within an organization. And, you know, I think sometimes technical team knows they can
[00:48:58] Paul Roetzer: do something, they go do it.[00:49:00]
[00:49:00] Paul Roetzer: Maybe the business team, the legal team isn’t even aware they’re doing it and you end up with terms of use that are, you know, misrepresenting things
[00:49:08] Paul Roetzer: to people. So, yeah, I unfortunately, you know, I think
[00:49:13] Paul Roetzer: LinkedIn got caught here doing something they shouldn’t have been doing based on the principles. And, um, again, I don’t
[00:49:21] Paul Roetzer: know though. I, I just assume. If everything’s public And it is what it is. Like they’re going to get trained on it. So, um, yeah, but if you don’t want it,
[00:49:32] Paul Roetzer: go and change the settings. We’ll put the link in there and you can follow through. It takes, you know,
[00:49:36] Paul Roetzer: 10 seconds or whatever to change the settings and turn it off if you want. But it doesn’t stop other people from
[00:49:40] Paul Roetzer: scraping the data and training their
[00:49:42] Paul Roetzer: models on it.
[00:49:43] NotebookLM
[00:49:43] Mike Kaput: Alright, next up, since
[00:49:46] Mike Kaput: we reported on Google’s Notebook LM last week and their new audio overview feature, we have seen this get a ton of buzz. as
[00:49:54] Mike Kaput: an extremely valuable and powerful AI tool that’s
[00:49:56] Mike Kaput: been flying pretty below the radar until recently.[00:50:00] And possibly in response to
[00:50:02] Mike Kaput: all these rave reviews coming out about NotebookLM,
[00:50:05] Mike Kaput: Google has officially launched it as what they call an additional service
[00:50:09] Mike Kaput: That basically expands its availability
[00:50:11] Mike Kaput: globally. It graduates it from what it used to be, which was an early access app. This is all
[00:50:17] Mike Kaput: just Google terminology. Basically
[00:50:19] Mike Kaput: an experimental app from Google. It is now a fully fledged service you can access. Now, as a reminder, Notebook LM uses advanced models like Gemini 1. 5 Pro to analyze, summarize, and converse with user uploaded source Documents and material. You can upload Google Docs, PDFs, website URLs, etc.
[00:50:40] Mike Kaput: And then you can start asking questions, generating
[00:50:43] Mike Kaput: summaries, and doing these really cool audio overviews based on those sources. So
[00:50:48] Mike Kaput: NotebookLM is basically a really powerful AI research assistant.
[00:50:53] Mike Kaput: So Paul, it’s kind of interesting here. It seems like
[00:50:56] Mike Kaput: Google has woken up to the fact NotebookLM is [00:51:00] pretty valuable and pretty popular.
[00:51:02] Mike Kaput: Like, have you Been exploring any additional use cases or projects or hearing anything more about it?
[00:51:07] Paul Roetzer: No, but I think what we got to do now, it’s
[00:51:09] Paul Roetzer: just a good reminder for people when you find these valuable use cases, you know, as an enterprise that has Google workspace to make sure the admin goes in and turns it on and that, and that
[00:51:18] Paul Roetzer: you provide training to people for how to get value from it. So, you know, you and I might talked about how
[00:51:23] Paul Roetzer: valuable we found that initially,
[00:51:25] Paul Roetzer: but we haven’t trained our team, how to use it.
[00:51:27] Paul Roetzer: We haven’t like laid out, here’s the five use cases of what
[00:51:29] Paul Roetzer: you should be testing and, you know, it’s turned on for you now. So just a reminder to people
[00:51:34] Paul Roetzer: To, you know, if you want to scale the value of AI, like it’s not just finding the cool tool and, you know, using it yourself, it’s okay, let’s take this to everybody else, who else on the team could benefit from this, the research team, the PR team, like
[00:51:48] Paul Roetzer: the, the blogging team, the podcasting team, like who could be using this tool and how much time could they be saving or how much output could they be increasing?
[00:51:56] Paul Roetzer: so, you know, think about that when you’re looking at these, these kinds of [00:52:00] kind of breakthroughs and capabilities, to really, you
[00:52:03] Paul Roetzer: know, play it out across your organization. It starts with who’s the admin that can turn this on in Google Workspace for us.
[00:52:10] Plex(dot)it
[00:52:10] Mike Kaput: so a fun follow up to a topic we covered last week, so,
[00:52:14] Mike Kaput: Paul, on September 15th, you posted on X that there was a need, that we all need a verb for what, what we call using
[00:52:23] Mike Kaput: perplexity, which is the AI powered search
[00:52:25] Mike Kaput: engine tool we all love. It’s becoming so popular that you kind of were saying it’d be helpful to have a way to tell someone to go use it. Just like we tell
[00:52:33] Mike Kaput: someone to go Google something, you suggested saying, quote, Plexit. Interestingly, Perplexity’s
[00:52:39] Mike Kaput: CEO posted something soon after, confirming that Plexit was the proper way to
[00:52:46] Mike Kaput: say this. Now, look, I’m not saying you came up with the term first, but this timing is
[00:52:51] Mike Kaput: a pretty big coincidence, I’m just saying. So take that what you will. Now, interestingly, Perplexity’s CEO posted again [00:53:00] that the company now owns the domain Plex. it, And if you type that in, it goes right to Perplexity.
[00:53:09] Mike Kaput: Paul, it certainly seems like Perplexity’s CEO is on board with your suggestion.
[00:53:13] Paul Roetzer: Yeah, yeah, it was funny. I mean,
[00:53:16] Paul Roetzer: I was going to let it go
[00:53:17] Paul Roetzer: and I thought, eh, whatever. I’ll have a little fun with it. yeah, it is. The timing is. perfect. Maybe coincidental, but yeah, I saw his tweet and I was like, wait a second. Did he go buy this
[00:53:28] Paul Roetzer: URL? And he, He totally did. And set up the redirect and, yeah, it’s a good for them.
[00:53:33] Paul Roetzer: If, if you want to do plex.it, it just redirects you to perplexity and may or may not have been influenced
[00:53:40] Paul Roetzer: by a tweet.
[00:53:41] Mike Kaput: Maybe if we say that often enough, Perplexity will use it as its training data. and start suggesting to people
[00:53:47] Paul Roetzer: who coined the
[00:53:48] Mike Kaput: yeah, there is some controversy.
[00:53:50] Paul Roetzer: dude, actually, I want to go into perplexity right now and ask who coined
[00:53:53] Mike Kaput: god, right?
[00:53:56] Introducing the ExecAI Newsletter
[00:53:56] Mike Kaput: All right, last but not least this week, Paul, [00:54:00] you just launched the first edition of what you’re calling the Exec AI Newsletter.
[00:54:05] Mike Kaput: And this is through Marketing AI Institute’s sister company, SmarterX. ai, which is an AI research
[00:54:11] Mike Kaput: and consulting firm. So the first edition just went out yesterday morning. Can you maybe tell us a little bit more about this newsletter, like who it’s for, why it’s needed?
[00:54:21] Mike Kaput: All for
[00:54:22] Paul Roetzer: business leaders.
[00:54:23] Paul Roetzer: And, you know, we looked at a lot of different things to do with this, Mike, like, you know, you and I talked about different paths to go. How do we make this different than the Institute newsletter, which is really like a This Week in AI, and
[00:54:33] Paul Roetzer: that you know, Summarizes what we talk about on the podcast. And so we decided to do with this one is, you know, again, focus on leaders
[00:54:39] Paul Roetzer: and emerging leaders and not just marketing, like the story beyond marketing is sort of the whole premise of SmarterX. and so we, we decided to
[00:54:47] Paul Roetzer: go with the path of an editorial from me each week, that’s specifically for leadership, and then a
[00:54:53] Paul Roetzer: presentation. Preview of what’s coming up on the podcast. So
[00:54:56] Paul Roetzer: I write this thing, usually Friday mornings is going to be the plan.[00:55:00]
[00:55:00] Paul Roetzer: It comes out on Sunday mornings, and it provides that kind of, you know, LinkedIn style, like two to three hundred words, maybe, intro editorial.
[00:55:08] Paul Roetzer: And then it’s just quick hitting bullets about what’s coming up, for the week, like the things we’re looking at.
[00:55:14] Paul Roetzer: And then interestingly, It’s a hundred percent human written. So I’ve made the choice not to use AI in the development of this. And I, somebody actually asked me about that, like on a LinkedIn comment, like, why wouldn’t you use
[00:55:25] Paul Roetzer: AI?
[00:55:26] Paul Roetzer: Just curious. And my whole perspective is like, because people, I want, I think people
[00:55:33] Paul Roetzer: who subscribe want to hear like my perspective and points of view. They don’t want something that anybody could generate in ChatGPT. And so for me,
[00:55:41] Paul Roetzer: it’s really important that I actually do the work. Like I think things through, I, I enjoy writing. So it’s like, I don’t really want AI to replace my writing.
[00:55:50] Paul Roetzer: it’s part of my daily process. And then it’s not
[00:55:54] Paul Roetzer: like that much more work for me to do it because I’m basically just writing about things that Mike [00:56:00] and I are prepping for anyway for the podcast.
[00:56:02] Paul Roetzer: And so it actually helps me kind of think through the topics we’re going to go through, so I don’t know, I think it’s
[00:56:07] Paul Roetzer: just a good reminder
[00:56:08] Paul Roetzer: for people, like, just because AI can do something doesn’t mean it should do it. That it’s okay
[00:56:13] Paul Roetzer: if there’s, you know, Human
[00:56:14] Paul Roetzer: things you want to hold on to because it’s what you do and what you find value in and because it creates value for other people.
[00:56:21] Paul Roetzer: And so is it the most efficient choice? No,
[00:56:24] Paul Roetzer: Like I could definitely write the newsletter faster if I was using perplexity and,
[00:56:29] Paul Roetzer: ChatGPT and other tools. But it’s really important to me that there is a very human element to this newsletter that I don’t want [00:56:37] ai [00:56:37] to do. and so
[00:56:39] Paul Roetzer: yeah, that, that’s kind of the story of it
[00:56:41] Paul Roetzer: is issue one dropped on Sunday and the plan is to do one each week and it’s a quick read.
[00:56:46] Paul Roetzer: It’s like, you know, three minutes if you read the whole thing, but hopefully it’s a value to people and, you know, gives people a little heads up about what what’s coming up in the week ahead.
[00:56:56] Mike Kaput: Awesome.
[00:56:58] Mike Kaput: Well, that’s all
[00:56:59] Mike Kaput: we got this [00:57:00] week, Paul. We covered quite a bit of ground. I appreciate, as always, you breaking everything down for us.
[00:57:05] Mike Kaput: just a couple quick
[00:57:06] Mike Kaput: final notes, as always, in terms of newsletters, we also have the Marketing AI Institute newsletter at marketingainstitute. com forward slash newsletter.
[00:57:16] Mike Kaput: It’s got all the
[00:57:18] Mike Kaput: things you need to know this week in AI.
[00:57:20] Mike Kaput: It’s a
[00:57:20] Mike Kaput: super easy and valuable way
[00:57:22] Mike Kaput: to keep up
[00:57:23] Mike Kaput: with everything in addition to
[00:57:24] Mike Kaput: the podcast. Also,
[00:57:26] Mike Kaput: if you can, leave us a review. We take the
[00:57:29] Mike Kaput: reviews very seriously and try to use them to improve, and they
[00:57:33] Mike Kaput: help us get the podcast to more people. Paul, thanks again.
[00:57:38] Paul Roetzer: Thanks, Mike. Talk to everyone next week. We’ll be back with a regular episode.
[00:57:43] Paul Roetzer: Thanks for listening to The AI Show. Visit MarketingAIInstitute. com to continue your AI learning journey, and join more than 60, 000 professionals and business leaders who have subscribed to the weekly newsletter, downloaded the AI blueprints, attended virtual [00:58:00] and in person events, taken our online AI courses, and engaged in the Slack community.
[00:58:06] Paul Roetzer: Until next time, stay curious and explore AI.